Martha Palmer

An enrichment of verb entries in WordNet that includes more specific syntactic information and verb class membership. It draws heavily on Levin Classes and the English Tree-Adjoining Grammar; (see also SIGLEX, and EAGLES, now ISLE)

(VerbNet also forms the basis of Parameterized Action Representations which are used for natural language interaction with virtual humans, work that is done at the Human Simultation and Modeling Center.)

We are using a similar approach to provide a companion lexicon for the Penn TreeBank. We created a Proposition Bank by tagging predicate-argument structures in the TreeBank with sense tags and argument labels based on lexicon entries with predicate-argument structure information. The choices we are making with respect to predicate-argument structure tagging also influenced the sense-tagging decisions in the preparation of training data for SENSEVAL2.

Korean/English Machine Translation, with CoGenTex and Systran, now at the completion of Phase II. An application of the dependency structure approach, as reflected in both Tree-Adjoining Grammars and Meaning Text Theory, to domain-specific machine translation. The components used in Phase 1 included the SuperTagger and the Collins parser. For Phase II we trained a new Korean morphological analyzer and postagger (Chunghye Han) and parser (Anoop Sarkar) and created a hand-crafted transfer lexicon (Chulwoo Park).

This project began with the annotation of 100K words of Xinhua newswire with syntactic structure, and now encompasses a 500K word Chinese Treebank (primarily Xinhua newsire and the Sinorama news magazine), as well as PropBank annotation, pronoun coreference annotation, word sense tags, and discourse connective annotation for the same data. A parallel 350K English corpus is also being annotated. Funding has been primarily from the ACE program, Darpa and NSF. Additional Broadcast news annotation is now being funded by Darpa.